User data is unlikely to add much to frontier AI’s math gains, one post argues

One post argues that frontier models’ gains in areas like math come from scaling pretraining and reinforcement learning with verifiable rewards, with user-data training exceedingly unlikely to contribute much. It suggests user data is more likely used to find failure modes or situations that hired annotators struggle to recreate, and calls for clearer disclosure of training practices. A related comment questions whether frontier labs would want Chegg’s user data, pointing skeptically to homework-solution data from struggling undergraduates.